pydantic
pydantic copied to clipboard
pydantic
Union of Annotated fields do not respect Field(strict=False)
Initial Checks
- [X] I confirm that I'm using Pydantic V2
Description
Cannot override strict mode for the annotated union of NewPath and Filepath fields when strict mode is set for all fields via ConfigDict.
Example Code
from pydantic import BaseModel, ConfigDict, Field, FilePath, NewPath
from typing_extensions import Annotated
class TestModel(BaseModel):
results_filepath: NewPath | FilePath
class StrictTestModel(BaseModel):
model_config = ConfigDict(strict=True)
results_filepath: (
Annotated[NewPath, Field(strict=False)]
| Annotated[FilePath, Field(strict=False)]
)
data = {'results_filepath': 'test.json'}
non_strict_model = TestModel(**data)
print(f'Non-strict model: {non_strict_model.model_dump()}\n')
strict_model = StrictTestModel(**data) # Exception raised
print(f'Strict model: {strict_model.model_dump()}')
"""
Output:
Non-strict model: {'results_filepath': WindowsPath('test.json')}
Traceback (most recent call last):
File "***\pydantic_test.py", line 23, in <module>
strict_model = StrictTestModel(**data)
^^^^^^^^^^^^^^^^^^^^^^^
File "***\venv_win3122\Lib\site-packages\pydantic\main.py", line 171, in __init__
self.__pydantic_validator__.validate_python(data, self_instance=self)
pydantic_core._pydantic_core.ValidationError: 2 validation errors for StrictTestModel
results_filepath.function-after[validate_new(), lax-or-strict[lax=union[json-or-python[json=function-after[path_validator(), str],python=is-instance[Path]],function-after[path_validator(), str]],strict=json-or-python[json=function-after[path_validator(), str],python=is-instance[Path]]]]
Input should be an instance of Path [type=is_instance_of, input_value='test.json', input_type=str]
For further information visit https://errors.pydantic.dev/2.6/v/is_instance_of
results_filepath.function-after[validate_file(), lax-or-strict[lax=union[json-or-python[json=function-after[path_validator(), str],python=is-instance[Path]],function-after[path_validator(), str]],strict=json-or-python[json=function-after[path_validator(), str],python=is-instance[Path]]]]
Input should be an instance of Path [type=is_instance_of, input_value='test.json', input_type=str]
For further information visit https://errors.pydantic.dev/2.6/v/is_instance_of
"""
Python, Pydantic & OS Version
pydantic version: 2.6.4
pydantic-core version: 2.16.3
pydantic-core build: profile=release pgo=true
install path: ***\venv_win3122\Lib\site-packages\pydantic
python version: 3.12.2 (tags/v3.12.2:6abddd9, Feb 6 2024, 21:26:36) [MSC v.1937 64 bit (AMD64)]
platform: Windows-10-10.0.19045-SP0
related packages: fastapi-0.110.0 pydantic-settings-2.2.1 typing_extensions-4.10.0
commit: unknown
@tropicoo,
Hmph, thanks for reporting this. Definitely a bug - can likely be fixed in the schema generation logic in pydantic. PRs welcome with a fix!
@sydney-runkle I can pick this up if no one is working on this. Can you please assign this to me?
The issue comes from the output of pydantic.plugin._schema_validator.create_schema_validator. create_schema_validator internally calls SchemaValidator from pydantic-core. The logic to build the validator in pydantic-core needs to be updated here
@sydney-runkle @samuelcolvin thoughts?
@SharathHuddar,
Sounds good, feel free to open a PR, I'd be happy to review!
@sydney-runkle I'm not able to figure out the right way to solve this. Please feel free to re-assign it to someone else
It seems that the error happens because of the union of annotated. Specifically, this has the problem:
class StrictTestModel(BaseModel):
model_config = ConfigDict(strict=True)
results_filepath: (
Annotated[NewPath, Field(strict=False)]
| Annotated[FilePath, Field(strict=False)]
)
but this doesn't:
class StrictTestModel(BaseModel):
model_config = ConfigDict(strict=True)
results_filepath: (
Annotated[NewPath, Field(strict=False)]
)
I think what's going on is that pydantic-core is not doing the right thing when determining the strictness to use within a union schema. There are a lot of ways this could be happening, I know @davidhewitt did some changes to union schema stuff a long time ago in the interest of trying to determine which case of the union was best-suited, and I wonder if the problem is it's trying to evaluate the inner schemas in strict mode under circumstances in which it should only try in lax mode. I think this is a significant step in the right direction of understanding what is going on BUT I wouldn't be surprised if it still takes a decent amount of work investigating inside pydantic-core to find the ultimate issue.
For what it's worth, @davidhewitt I think there's a chance you'd be able to figure this out quickly.
Hey,
Indeed, the field's strict parameter is not respected when validating union type. If I'm not incorrect, it seems that UnionValidator only uses its "global" strict_mode (from ValidationState) instead of what's encoded into a field?
https://github.com/pydantic/pydantic-core/blob/34d789fc510810eee507347d5e0897d8fb9045bb/src/validators/union.rs#L123
https://github.com/pydantic/pydantic-core/blob/34d789fc510810eee507347d5e0897d8fb9045bb/src/validators/model_fields.rs#L123
Test examples:
Here are a couple of test examples with ModelValidator
This does not work:
def test_model_field_with_union_type_and_strict_mode_disabled():
class MyModel:
__slots__ = (
"__dict__",
"__pydantic_fields_set__",
"__pydantic_extra__",
"__pydantic_private__",
)
v = SchemaValidator(
{
"type": "model",
"cls": MyModel,
"schema": {
"type": "union",
"strict": True,
"choices": [
{
"type": "model-fields",
"fields": {
"foo": {
"type": "model-field",
"strict": False,
"schema": {"type": "int"},
}
},
},
{
"type": "model-fields",
"fields": {
"bar": {
"type": "model-field",
"strict": False,
"schema": {"type": "int"},
}
},
},
],
},
}
)
m = v.validate_python({"foo": "123"})
Validator:
ModelValidator {
revalidate: Never,
validator: Union(
UnionValidator {
mode: Smart,
choices: [
(
ModelFields(
ModelFieldsValidator {
fields: [
Field {
name: "foo",
lookup_key: Simple {
key: "foo",
py_key: Py(
0x00007f7d48d805d0,
),
path: LookupPath(
[
S(
"foo",
Py(
0x00007f7d48d82d90,
),
),
],
),
},
name_py: Py(
0x00007f7d4bc194a0,
),
validator: Int(
IntValidator {
strict: false,
},
),
frozen: false,
},
],
model_name: "Model",
extra_behavior: Ignore,
extras_validator: None,
strict: false,
from_attributes: false,
loc_by_alias: true,
},
),
None,
),
(
ModelFields(
ModelFieldsValidator {
fields: [
Field {
name: "bar",
lookup_key: Simple {
key: "bar",
py_key: Py(
0x00007f7d48d80630,
),
path: LookupPath(
[
S(
"bar",
Py(
0x00007f7d48d810e0,
),
),
],
),
},
name_py: Py(
0x00007f7d4bc19440,
),
validator: Int(
IntValidator {
strict: false,
},
),
frozen: false,
},
],
model_name: "Model",
extra_behavior: Ignore,
extras_validator: None,
strict: false,
from_attributes: false,
loc_by_alias: true,
},
),
None,
),
],
custom_error: None,
strict: true,
name: "union[model-fields,model-fields]",
},
),
class: Py(
0x00005561e20c1fd0,
),
post_init: None,
frozen: false,
custom_init: false,
root_model: false,
undefined: Py(
0x00007f7d4c5c13a0,
),
name: "MyModel",
}
This works:
def test_model_field():
class MyModel:
__slots__ = (
"__dict__",
"__pydantic_fields_set__",
"__pydantic_extra__",
"__pydantic_private__",
)
v = SchemaValidator(
{
"type": "model",
"cls": MyModel,
"schema": {
"type": "union",
"strict": True,
"choices": [
{
"type": "model-fields",
"fields": {
"foo": {
"type": "model-field",
"strict": False,
"schema": {"type": "int"},
}
},
},
],
},
}
)
m = v.validate_python({"foo": "123"})
Validator:
ModelValidator {
revalidate: Never,
validator: ModelFields(
ModelFieldsValidator {
fields: [
Field {
name: "foo",
lookup_key: Simple {
key: "foo",
py_key: Py(
0x00007fd5b5b191d0,
),
path: LookupPath(
[
S(
"foo",
Py(
0x00007fd5b72a1f20,
),
),
],
),
},
name_py: Py(
0x00007fd5b83154a0,
),
validator: Int(
IntValidator {
strict: false,
},
),
frozen: false,
},
],
model_name: "Model",
extra_behavior: Ignore,
extras_validator: None,
strict: false,
from_attributes: false,
loc_by_alias: true,
},
),
class: Py(
0x00005593d66b1880,
),
post_init: None,
frozen: false,
custom_init: false,
root_model: false,
undefined: Py(
0x00007fd5b8cbd3a0,
),
name: "MyModel",
}
@mikeleppane,
Great analysis here. Will definitely be helpful to have if we move forward with this as a desired change!
Hmm, this still needs some internal discussion with the team re strict behavior for unions, will bring this up at our sync tomorrow.